Hi Team, we have been observing the huge delays in the query response times and broker CPU spikes up and the historical CPU is being stable. we mostly use joins in almost all the queries with CTE on druid. Is the CTE on druid pulling the records from historicals and performing the joins on the brokers. ? Can someone please suggest anything specific that i can look to efficient run the queries ?

Hi Sai, can you share an example query? Generally final query processing is done on the Broker ... but joins should be able to be done via broadcasting secondary tables to the compute nodes for joining first, and those tables can be filtered. There may also be ways to consolidate extra joins. But I couldn't guess what optimizations might be available without seeing the actual queries. Thanks. John